Classification of UAVs using Time-Frequency Analysis of Remote Control Signals and CNN

Abstract

Unmanned aerial vehicles (UAVs) have re-cently gained a significant interest in the research com-munity owing to their unrivaled commercial chances in wireless communications, search and rescue, surveillance, logistics, delivery, and intelligent agriculture. In safety-critical applications such as intrusions, identifying the type of drone enhances the countermeasures. This paper proposes classifying UAVs from radio frequency (RF) fingerprints using time-frequency transformation and con-volutional neural networks (CNN). The proposed method-ology involves RF fingerprints' wavelet synchrosqueezed transform (WSST) followed by a proposed lightweight CNN model. The methodology is verified on a data set containing fifteen different classes of drone's RF fingerprint. The proposed CNN model size, Raspberry Pi deployment feasibility, and accuracy are compared with the existing pre-trained state-of-art deep learning models. The proposed model achieves a testing accuracy of 99.09% at 387 kilobytes (KB) size and can run on Raspberry Pi in 25.54 milliseconds.